Good morning.
Futures are quiet but the real tape is bonds — the Bessent bid is fading, long-end yields creeping back toward that 5.30% ceiling. JPMorgan's verdict: the buyback "addresses the symptoms rather than the cause"; fiscal-Fed fissure still open. Don't fight the slow bleed into AI financing costs.
Overnight the big print: BABA PROFIT -75% on AI spend plus soft China consumption — strategic capex, not value destruction, IF cloud growth reaccelerates. Pop Mart warning on a 20% sales growth miss says the consumption side is still leaking. Anthropic noise keeps building: RUN-RATE $65B, Q2 REVENUE $11.6B, and SemiAnalysis says July's ARR "miss" is an IPO accounting switch (run-rate → annualized monthly), not demand stalling. Don't trust the wobble either way — S-1 is the only judge.
Asia: JAPAN SEMI EQUIP EXPORTS +49.1% YoY, fifth straight month; global server 2Q shipments missed ONLY because memory/CPU/IC substrate supply choked. That's supply, not demand. Samsung and SK Hynix record shareholder returns into the selloff = a floor, not a reversal.
Three themes: (1) the binding constraint is physical — equipment, memory, power — and the equipment chain (AMAT, LRCX, KLAC, ASML) owns the pricing power. (2) China's model layer sells at ~1/4 of Claude's cost; BABA's profit sacrifice is the entry fee, and US model-layer pricing anchors to that — enterprise lock-in is the only defense. (3) Power is the new queue: Denmark puts data centers last in the grid line, and off-grid turbines are already cracking (xAI, Crusoe, Vantage all reported failures). Hyperscalers with locked-in power win; everyone else pays the reliability tax.
We'll hit up MRVL and CRWD first, then get to the memory/capex complex — SSNLF, MU, AMAT.
Google finally gives MRVL a third hyperscaler to hang its ASIC story on — that's real, and the market's right to cheer. But read the fine print before you chase it: this is NOT a core TPU accelerator win. It's the attach silicon around the TPU, and the $120B headline is a maximum-scale framework with zero purchase commitment. Stock at $237, up 234% in a year. The easy money went in before this print.
Six firms, all Buy/Overweight-equivalent. PT range: $225 (TD Cowen, raised from $200) to $350 (Stifel). The cluster sits at $235-275 — Raymond James $235, JPMorgan $240, Barclays $275 — with UBS at $310 (raised from $300) and Stifel at $350 as the true believers in the ceiling.
The math that matters:
MRVL JUST LANDED A THIRD HYPERSCALER. Raymond James — which built its ASIC thesis on Amazon and Microsoft only — admits it:
"Google had not been incorporated into the firm's assumptions prior to this deal." — Raymond James
That's the true incremental. The known narrative — AMZN/MSFT ASIC ramps, optical DSP share resilience, the $1B CXL attach revenue guide for CY2028 — just got a new pillar.
But the "expanded partnership" framing does a lot of heavy lifting. JPMorgan's Harlan Sur was explicit: this is XPU attach and TPU-adjacent products — AI inference offload engine (SRAM-based), storage controllers, NIC/SmartNIC/DPU controllers, CXL memory interface controllers. Barclays draws the same line:
"The project is likely a Google effort for a more tailored inference solution using Marvell intellectual property, rather than a tensor processing unit project." — Barclays
Broadcom still owns the core TPU volume under its long-term supply agreement. And on the headline number, UBS is blunt:
"A maximum-scale framework rather than committed revenue or a forecast." — UBS
Backend-loaded, too. UBS only moved FY28 revenue to $25.9B from $24.4B and EPS to $10.31 from $9.82 — the big money lands post-2028, post-design-cycle.
Bull: A third hyperscaler validates the XPU attach thesis, and this one carries a $120B ceiling — even 25-30% realization is a needle-mover. The warrant structure aligns Google's equity upside directly with silicon spend; no purchase commitment, but Google only gets paid if it buys. UBS sees $15 EPS by 2030. The CXL memory interface work here is the ecosystem anchor for MRVL's $1B CXL attach guide.
Bear: This is the silicon NEXT TO the TPU, not the TPU — Broadcom owns that. No formal purchase commitment anywhere, and the deal is backend-loaded to the point where UBS's FY28 raise was only ~4%. JPM's $240 PT sits 1% above spot. The 58.97M share warrant is dilution optics if the deal ramps, and after a 234% run, new money at $237 gets no margin of safety.
Hyperscaler custom silicon is officially a supply-chain diversification theme, not a Broadcom monopoly. Google going MRVL for TPU-adjacent inference and CXL work is a data point for AVGO's hold on the core and a validation of ALAB's addressable model. The warrant-for-spend structure is the sleeper here — expect it to become a template for custom silicon deals. That changes how you model dilution and alignment across the whole ASIC complex. Watch the CXL angle specifically: memory interface controllers are the quiet bet, and MRVL just got the most credible inference anchor that product line could ask for.
Verdict: The analog upcycle has legs, and the sell-side is catching up. Bernstein flipped ADI to Outperform, Needham lifted PT to $450 — two independent routes to the same call: this cycle runs longer and gross margins keep grinding higher.
Q3 was a beat in every lane. EPS $3.45 vs $3.34 est., revenue $4.02B vs $3.91B est. — FIRST $4B QUARTER EVER, UP 40% YOY. The $110M revenue beat matters at this scale. Needham says Q4 guidance came in above their model too. Gross margin steps UP to ~74% in Q4 from 72.5% in Q3, driven by mix plus September price increases — that's pricing power landing in the P&L, not just a mix shift.
Management's framing: AI and defense spending stretch this cycle past the typical duration. Inventories stay lean, restocking hasn't really started. That's the bull case in one sentence.
Bernstein's upgrade is valuation math, plain and simple. Stock trades low-20s on FY27 EPS, PEG 0.67, and $20 EPS by 2028 is "plausible." They did cut the multiple — 25x from 28x — but raised estimates enough to net a $35 PT increase. Honest work.
"Valuation may have become more decoupled from growth, with risk and reward skewing positive."
That's the whole argument in one line. The same note airs the bear case: industrial growth runs +50% YoY and some PMs read that as a peak, not a plateau. Factory-utilization GM tailwinds don't recur.
Bull: Upcycle with a runway. Optical revenue doubles in FY26, industrial recovers hard, September pricing hits. Forward estimates converge near $18.30-18.85 (Needham's CY28 $18.83, Bernstein's FY27/28 avg $18.34). 24-25x on that gets you $450-465. The math works.
Bear: Stock's up 54-66% in a year. Multiple expansion is done — even the bulls anchored to 24-25x, down from Bernstein's prior 28x. From here it's all delivery on high-teens EPS. Industrial rolls over and you lose the GM story and the cycle thesis at the same time. Nobody's underwriting a bigger re-rating.
Bottom line: ADI is a Buy, but the easy money's been made. $450-465 is the new band. Add on strength, not on hope.
UBS, Mizuho, Needham, Rosenblatt all stayed long after the Supernova event — NO ONE BLINKED. The tape disagrees: stock down 16% last week, 31% over six months, now at $202.10. But this is a product story, and the product is the real deal. The margin debate is where the r/r actually lives.
The CS-4 is the headline. 6x system-level improvement vs the CS-3 because Cerebras crammed THREE wafer-scale engines per system instead of one. Per-WSE token performance roughly 2x. Latency down 2.5x. OpenAI's GPT-5.6 Sol runs 14x faster on Cerebras than GPU systems — and the inference gap stretches to 30x.
The Nexus rack platform matters more long-term. It's modular, spans CS-5 and CS-6, with CS-5 targeting 20x throughput and 4x tokens/sec by 2027. Hot Chips on Aug 25 is the next catalyst for architecture detail.
The full buy-side crew is sticking. PTs cluster at $300-330 — UBS at $330, Needham and Rosenblatt at $300, Mizuho trimmed to $300 from $310. Nine analysts revised estimates upward recently. That's not panic positioning.
Mizuho's margin point is the key nuance: FLAT BILL-OF-MATERIALS COSTS across generations plus a 20x performance uplift equals pricing power. Same BOM, 20x the throughput — that's how you expand hardware margins.
"UBS views the CS-4 launch as the highlight of the event and continues to view the recent pullback as a compelling buying opportunity."
REVENUE $209.9M — a record, +103% YoY, +10% QoQ. September guide at $215M vs $216M consensus — basically a rounding error. FY26 guidance above Street. Management says capacity expansion supports 3x+ revenue growth in FY27.
The bottom line is a mess, though. Mizuho cites a $0.05 per-share loss. UBS prints ADJUSTED EPS AT -$2.98 VS -$0.18 CONSENSUS. Those are not the same number. GAAP vs adjusted, and the gap smells like stock comp or a one-time charge — nobody's explaining it, and PMs should ask before adding.
Bull: This is a real architectural step-change, not a spec bump. Three WSEs per rack, flat BOM, 30x inference advantage, OpenAI running on it, and Mizuho flags Meta as a POTENTIAL NEW CUSTOMER in 2H26. At $202 you're paying for the roadmap — and the roadmap is credible.
Bear: The stock has been a serial disappointment — down 31% in six months. Mizuho CUT the PT on margin pressure, which undercuts the flat-BOM bull case at the margin. The adjusted EPS print is horrid no matter how you adjust it. And CS-5 doesn't land until 2027. Long time to carry a ~$51B market cap on a promise.
Stock at $202.10, down 16% on the week. Supernova didn't change the thesis — it validated it. The question is whether the market starts pricing 2027 revenue off the CS-4 ramp or keeps treating CBRS like a GPU also-ran with a neat architecture. Hot Chips on Aug 25 is the next check.
VERDICT: STOCK'S UP 272% IN A YEAR AND TRADING ABOVE THE NEUTRAL GUY'S PRICE TARGET. THE DEBATE ISN'T WHETHER VOICE AI IS REAL — IT'S WHETHER ALL THE GOOD NEWS IS ALREADY IN THE TAPE. Needham says keep running, $60. Piper says risk/reward stinks, $52. B.Riley says you're all too small, $85. When the spread between targets is 60%+, the market's telling you this is a story stock now.
Needham came away from investor meetings with CEO David Morken and CFO Daryl Raiford and struck a confident tone. Demand pipeline up YoY. Revenue build reflects a $1M+ CUSTOMER RAMP TRAJECTORY — the whales are scaling, not just signing. Political revenue ests of $13M for 2026. They model a 20%-PLUS ENTERPRISE VOICE EXIT RATE IN Q4 2026. That's the re-acceleration story, quantified.
"The demand environment and pipeline for the company is up year-over-year. Current expectations and estimates are positioned to beat moving forward."
Piper Sandler is the wet blanket — and it's honest about why. Raised PT to $52 from $38 (Neutral), but BAND's already at $53. They estimate the voice-only slice of Agentforce CX adds 2-3PPS TO GROWTH in 2027-28. Real, but not the kind of number that justifies a 272% year. They also flag the gross-impact problem: BAND counts many leading CCaaS providers as customers, and those are exactly the guys Salesforce's Agentforce could squash. If your biggest customers get disintermediated, your revenue gets a haircut.
B.Riley is still the bull's bull at $85. They give management credit for the Q2 print — revenue, EBITDA, EPS all modestly above. Core Cloud Comms grew 11.8% YoY. Full-year guidance raised on revenue and EBITDA. (They also trimmed EPS guidance for planned infra investments — that's the one blemish in the bull case.)
Bull: This is a beat-and-raise loop with a new growth engine bolted on. Core business re-accelerating (22% total revenue growth last quarter), Agentforce adds an upside kicker starting 2027, political year tailwind in 2026, and a cleaner cash flow margin profile going forward. Needham's right that estimates look positioned to beat — the setup into year-end is favorable.
Bear: You're paying a 272%-appreciation multiple for a 2-3pp Agentforce kicker that lands two years out. Piper's caution about growth rates beyond this year is the real risk: the reported numbers decelerate once the even-year comps and political revenue anniversary. Plus the CCaaS customer overlap is a structural vulnerability, not a footnote. At $53, you're buying the story at peak narrative.
Two raises, zero drama. Jefferies to $17.70, BofA to $21.40 — both Buy, both anchored on the same narrative: AI revenue hit 56% of mix and the operating leverage is finally showing up in the P&L. Stock trades at $12.05. BofA sees 85% upside; Jefferies is more conservative but still ~47% above the tape. Not a disagreement — just a r/r debate on the multiple.
Q2 was clean. Revenue RMB3.07B, +31% YoY, 1% above consensus. Adjusted EBITDA RMB1.1B, +171% YoY, 22% above consensus. Non-GAAP net loss narrowed 80% to RMB60M. And the headline: FIRST-EVER POSITIVE GAAP OPERATING PROFIT — RMB23M. That's the number that changes the conversation.
Gross margin recovered 2.4pts QoQ to 15.2%. Still weak by absolute standards (TTM sits at ~14.9%), but the rate of change is the trade. Both desks expect top-line acceleration in 2H on AI cloud demand.
"BofA sees potential top-line growth acceleration in the third and fourth quarters of 2026 driven by AI cloud demand."
MaaS is early innings — that's the optionality, not the current revenue driver.
Bull: AI revenue is 56% of the mix and still growing faster than the base. EBITDA +171% YoY. If gross margin keeps recovering 2+ points per quarter, GAAP profitability is no longer a question mark — it's a trajectory. At 3x forward sales (BofA's basis), a company inflecting to sustained profitability rerates quickly.
Bear: Gross margins are still 15% in a market where hyperscalers set the price. Non-GAAP net loss is narrower but still negative. And the stock has already re-rated hard into this print — $17.70 vs $21.40 PT spread tells you even the bulls can't agree on the multiple. MaaS could burn cash before it contributes.
Net: the operating leverage is real, not narrative. First GAAP op profit is an inflection. The debate from here is multiple expansion, not whether the story works. Decent hold-through-earnings name if you want a smaller, higher-beta expression on the AI cloud theme — just don't confuse a narrowing loss with a winner yet.
Margins are the story. Growth is the worry. Two Buys this morning — Needham at $75, UBS raised to $60 from $55 — and they're not buying the top line. They're buying the margin guide.
THE F2027 EBIT MARGIN EXPANSION GUIDE CAME IN AT >400BPS, STREET WAS AT ~360. That's the headline. Core revenue guide of ~13% YoY? Modestly below the ~14% Street bogey. The market is being asked to look past deceleration toward profitability. That's the trade.
Q4 print was clean: ADJUSTED EPS $0.84 vs $0.70 EST. REVENUE $436.2M vs $430.37M EST. Core revenue growth ~16% YoY. Management credited stronger margins, the restructuring, and AI product adoption. The guide is the give-back: ~13% core growth next year.
UBS values this at ~11x CY27 EV/FCF with >20% FCF growth. That's the r/r in one line. If margins inflect and FCF compounds, the multiple is cheap. If growth keeps fading, 11x is fair for a decelerator.
The real narrative change underneath all of this: Bill.com is deliberately shifting to higher-quality customers. Fewer total customers, higher TPV, lower take rate. It's a gross-profit-positive trade but it wrecks the optics. Net new adds were the weakest in a couple years — 1,800 AP/AR, 900 S&E. The AP/AR take rate slipped to 0.16% and missed guidance. The bears will hang their hats on that.
The accounting presentation change on rewards is the quiet tell. It makes the model comparable to other software companies and surfaces the profitability improvement from the RIF. That's a deliberate narrative reset. Needham's framing:
The accounting change related to rewards will make the company's model more comparable to other software companies and highlight a profitability improvement following a recent reduction in force.
Bull: Margin inflection + Rule of 40 by end of F2027 + 20% FCF growth at ~11x EV/FCF = re-rating fuel. Needham says the stock at $47.71 is undervalued. They see the strategy shift as favorable to gross profit.
Bear: The growth engine is sputtering. Weakest customer adds in two years, take rate below guidance, core revenue guide under Street. This "higher quality customer" pivot is a polite way of saying they gave up on volume. If the margin story disappoints even once, the multiple de-rates fast.
One flag from Needham: the company needs more Ad Valorem traction to maximize the stock. That's the unlock beyond basic expense discipline. Without it, this is a cost-cut story, not a growth story.
Two Buys, $60–$75 PTs, stock at $47.71. The Street's not fully paying for the margin guide yet — and the net customer adds tell you why. The bull case works if you trust the FCF math and the Rule of 40 trajectory. The bear case works if you think the top-line step-down is structural, not strategic.
Watch the take rate stabilization and the quarterly add prints. Those are the bogeys. Margins get you the re-rate. Growth gets you the multiple. Right now, BILL is only offering the former.
Verdict: The stock's already sprinted 90% in six months — now it needs to justify the multiple. Guggenheim's the latest to bump (PT to $162 from $138, Buy), but the tone is cautious: expect a Q2 beat, don't expect a guide raise. The market's pricing in the AI identity narrative; the actual print Wednesday is a lower bar than the guidance commentary.
Bull case writes itself: identity is the choke-point for AI security spend. Wells Fargo went Overweight with a $180 PT on AI-driven growth; Citizens upgraded to Market Outperform at $170 on "identity as critical infrastructure." Bernstein's survey shows cyber budgets growing faster than IT budgets; Goldman sees AI security spend inflecting late-2026/early-2027. That's a nice tailwind narrative with macro support.
Bear case is mean-reversion math. At $141, the stock's already lapped most of these PT hikes (Wells at $180, Citizens at $170, Stifel at $160, Guggenheim at $162 — that's an $18 spread, and the low end is only 15% above spot). Guggenheim's field checks are MIXED: 2 partners beat, 1 met, 2 missed — though all hit pipeline expectations. Federal ACV ran $12.2M, roughly 15% of New ACV, seasonally hot but lumpy.
The real bogey is guidance. Guggenheim sees FY27 total revenue guide of 9-10% as attainable, but:
"does not see much room for an increase."
That's the worry. Stifel's modeling 2% revenue upside and 2.5-3% cRPO upside — fine. But if Q3 guidance comes in at or below street, a multiple that's already stretched gets tested. The beat is likely. The guide is the trade. Reports Wednesday after close.
UBS raises to $390 from $300 — and STAYS NEUTRAL. That's the whole story in one line. They're rolling to CY28 at 46x FCF (was 42x CY27), but the stock is UP ~140% SINCE THE APRIL MYTHOS PRINT. At 52x EV/FCF on CY27 against ~15% pro forma revenue growth, you're paying a premium multiple for a growth rate that hasn't re-rated.
The bulls have their own math. Cantor ($425), Stifel ($415), Citizens ($415), TD Cowen ($400) — all Buy/OW. Prisma AIRS traction is real per channel checks, partner program expanding globally. Nobody's shorting the AI security narrative.
But UBS's checks are the honest tell: "did not observe a material uptick in tone." Steady demand, not accelerating. That's the difference between holding and chasing at 52x.
"We see a somewhat challenging setup heading into Palo Alto's fiscal year 2026 year end given the stock's recent performance."
Hard to argue with that after a 148% six-month run. Great company, rich price, crowded trade. I'd want a pullback for a new entry — not a fresh add up here.
TD Cowen bumps SNOW to $370 from $300, staying at Buy, on bullish channel checks into the September 2 print. Stock's already up 48% YTD at ~$325, knocking on the 52-week high of $341.95. This isn't a tired "AI is coming" story anymore — the checks say CoCo, the AI coding tool launched in February, is showing real usage velocity, and Cortex AI Gateway (released last month) is another potential kicker.
The collective analyst picture is loud and getting louder. UBS at $425, Citizens at $408, Stifel at $350 with a "significant QoQ increase in CoCo usage" — this is a tape where momentum begets momentum. 31 analysts revising FY estimates upward into the print. Street's looking for ~36-37% revenue growth in Q2, with FY27 in the high-30s to 40% range. TD Cowen sees ANOTHER QUARTER OF GROWTH ACCELERATION in Q2, which would be the second straight. The valuation pressure keeps building.
"The introduction of the CoCo AI coding tool in February is driving multiple new AI accelerants for the company."
Not everything is clean. Morgan Stanley's still at $300 — below the current price — worried about mixed AI adoption beyond the core data warehouse. That's the honest risk: CoCo is exciting, but the 36-37% growth bogey is high, and the stock's already priced a lot of good news. If the Q2 guide doesn't accelerate, the air pocket gets real. But right now, the check data and the buyside positioning say the path of least resistance is up into the print.
Barclays keeps the faith — Overweight, $300 PT — and the bull case writes itself. HBM leadership is the entire story, and nobody's mounting a serious challenge to that crown.
The kicker isn't just the memory upcycle. It's the capital return math: BARCLAYS SEES ~15% OF MARKET CAP RETURNED while the company still expands capacity and funds new opportunities. GROSS MARGIN 70.2%, REVENUE +145% LTM. That's a compounding machine, not a cyclical trade.
The rest of the street is lining up on the same side — fresh initiations from Wolfe (Outperform), Rosenblatt (Buy), Needham (Buy), and Goldman holding Buy. Collective thesis: memory tightness persists, ASPs hold even as hyperscalers trim content to manage constraints, and SKHY is the cleanest HBM expression in the sector.
"Barclays expects SK Hynix to remain the leader in high-bandwidth memory and prefers exposure to HBM within the memory sector."
Real catalyst is Q3 earnings — that's when we get the shareholder return framework and cap-ex intensity in hard numbers. PT spread is wide ($152 to $355), so there's genuine debate on durability. But at 18.3x with this momentum, Barclays sees the r/r as compelling. Hard to argue with that from the long side.
Truist is the bull to watch here — reiterating Buy at $155 after hosting CoreWeave IR for an investor call (100+ registered, so Street engagement is real). The whole thesis hinges on re-contracting: CoreWeave's already printing 67% gross margins, and Truist sees structural upside as legacy contracts roll into premium-priced renewals. The evidence anchor? Recent SpaceX compute deals pricing at a premium.
The re-contracting opportunity is a potential long-term structural tailwind to margins.
Managed inference is the emerging growth driver — it moves CoreWeave up the stack and into direct competition with pure inference cloud providers. That's a new competitive dynamic, but also a higher-value revenue mix. Worth watching how that changes the customer conversation.
Q2 was clean: revenue beat by 0.8%, adjusted operating income +82.9%, and 450 MW energized — that alone exceeds the nearest competitor's total capacity.
PT dispersion tells you everything about the debate: Cantor at $178, Truist at $155, Piper at $153 on the bull side; DA Davidson neutral at $100; Bernstein Underperform at $74. That's a 2.4x spread. The bull case is capacity lead + re-contracting compounding. The bear case — per Bernstein — is that the capex treadmill and competitive intensity eat the margin gains. Not sure we can read much into the 0.8% revenue beat; that's noise. The margin trajectory is the signal.
Rosenblatt starts coverage with a Buy and $410 PT vs $344.72 close — ~19% upside, and they're not being cute about it. The core thesis: the search-panic de-rating is DONE, and GOOGL is now being paid as an AI infrastructure owner. At 17.3x P/E with a 0.15 PEG and 20% revenue growth, the market is still pricing this like a mature search monopoly, not a hyperscaler at the center of the capex supercycle.
"The re-rating from search concerns has largely played out and Alphabet is now being compensated for its role in AI infrastructure."
Cloud is the swing factor, not search. Rosenblatt sees cloud sustaining momentum and driving multiple expansion as operating profit scales ahead of expectations — with TPU monetization kicking in past 2027. That's the bull case in one line: capacity + custom silicon = margin upside the street hasn't modeled. The bear case isn't a new search threat — it's cloud deceleration. That's the single metric to watch.
Supporting signals: the Marvell custom silicon deal (up to $120B cumulative revenue over ~6.5 years, warrants at $206.58) validates the TPU/compute trajectory, and Morgan Stanley flags big 2027 capex increases across the group. Meanwhile the YouTube-NBA streaming hub ($1B+, 25 teams local broadcasts from 2027-28) is a nice non-search call option nobody's paying for. Spirit Airlines data win for $10M is noise, but shows the data monetization muscle. Net net: the risk/reward is skewed long cloud, with search skepticism already in the price.
GUGGENHEIM RAISES PT TO $28 FROM $24, BUY MAINTAINED. Stock at $21.32 — that's ~31% upside into Thursday's Q2 FY27 print. Setup reads like a beat-and-raise: modest revenue upside, ARR just in line, and a Q3 guide that should clear the Street.
The narrative hinges on H2, not Q2. Q1 revenue came in range but short of consensus — management blamed a back-end-loaded quarter and larger deal timing. The 45/55 H1/H2 net new ARR split (vs 41/59 in FY26, 42/58 in FY25) is the tell. They're front-loading the narrative now so the H2 print hits harder.
Guggenheim sees S potentially nudging FY27 revenue guidance above the $1.195B-$1.205B range — that's 20% YoY growth. The company has a two-year cadence of modest beats, so the bar is low and they keep clearing it. Valuation is reasonable: 5.1x EV/NTM recurring revenue with 20%+ growth and 73% gross margins. Analysts have S turning profitable this year.
"Guggenheim believes SentinelOne can guide third-quarter revenue at least in line, if not modestly above the Street and perhaps modestly raise fiscal 2027 revenue guidance."
Guggenheim bumps VEEV to $276 from $232 into the Aug 26 print, keeping Buy. This is a slow-burn beat story, not a blowout: VEEV at least meets Q2, guides Q3 roughly in line, then lets upside compound as 2H comps ease. The real target is upstream — Guggenheim models FY27 NORMALIZED BILLINGS AT 15% YOY GROWTH, ~200BPS ABOVE CONSENSUS, and that outer-year forecast drives the DCF bump.
Customer conversations on AI positioning are positive, which matters when you're paying 44x P/E and 2.2 PEG. Stifel sits at $275 (subscription growth beats guidance), Oppenheimer at $225 (AI labor augmentation) — the bull camp's converging in the $225-276 zone around one narrative: Veeva's transition from Salesforce tenant to life sciences platform competitor keeps gaining traction, and Q2 is the next proof point.
"The firm sees more upside to fiscal 2027 normalized billings, as its plausible model yields 15% year-over-year growth, or about 200 basis points above consensus."
Bitcoin miner story is dead. Long live the HPC story. Rosenblatt bumps HIVE to $6 from $5.50 (13x FY28 EBITDA) and the stock sits at $2.82 — a $770M market cap against a target range of $4.50 to $10. That's a 2x-3.5x gap, and the re-rating catalyst is finally showing up in the financials.
FQ1 revenue +73% Y/Y to $79.1M, HPC revenue beat Rosenblatt's number for the first time in a year. Mining margins +500bps on firmware optimization — that's the kind of operational detail that actually matters when BTC price assumptions are being cut. But the headline GAAP print is horrid: -$142.9M net loss, $84.7M of it a non-cash Swedish tax provision. Adjusted EBITDA flipped to +$13.4M from -$9M prior quarter. The trajectory is the story.
The linchpin is the $350M, five-year AI cloud services deal with an investment-grade enterprise customer. Adds $70M in incremental annual recurring revenue, pushes total contracted ARR to ~$180M. Not sure we can read too much into a single contract, but it validates the thesis that HIVE isn't just a bitcoin-levered proxy anymore.
"The firm noted that Hive Digital's fiscal first-quarter 2027 results came in below consensus. Mining margins improved 500 basis points despite a challenging environment due to firmware optimization, and high-performance computing revenues beat the firm's estimate for the first time in a year."
Execution and funding risk remain elevated — that's the honest caveat here. FY28 estimates rise significantly, but near-term is dragged by lower BTC assumptions. H.C. Wainwright still at $7, so there's a $1 spread between buys. High volatility crew (beta 3.7), no profitability expected this year. The r/r is asymmetric but only if you can stomach the swings. HPC momentum accelerates, and the market is still pricing this like a miner.
Guggenheim bumps PT to $277 from $245, stays Buy — and the whole thesis is one number: the Street is modeling a ~70% adjusted New ACV decline in FY27. That is not happening. Autodesk's transition noise (reseller-to-direct mix) is getting confused with demand destruction, and the opportunity is in that misread.
The beat story is a back-end-loaded one. Q2 plausible case lands at 15.6% total revenue growth, ~120bps above consensus — actually below the 160bps average beat of the last eight quarters. So don't expect fireworks on August 27. The upside expands from there: Q3 at ~240bps above, FY27 at ~200bps above revenue and 640bps above on billings.
"Street numbers imply an adjusted New ACV decline of roughly 70% year-over-year, which the firm views as unlikely."
That's the whole ballgame. If the ACV cliff is a mirage, FY27 estimates are too low and the multiple gets interesting. Fundamentals back it — 92% gross margin, 18% revenue growth LTM.
The rest of the Street is already in the same zip code: Barclays $300 OW, Stifel $285 Buy (mixed channel checks), KeyBanc $341 OW (same direct-sales shift read). Goldman is the lone Neutral at $260. So the bull case is crowded-ish, but the delta from here is whether the ACV "decline" inflects cleaner than the whisper numbers imply. Positioned for that. 92% gross margins make this a quality compounder when the noise clears.
BofA cuts PT to $550 from $660, keeps Buy. The tell isn't demand — it's optics. Demand stays robust but supply constraints cap FY26 revenue upside, so the whole optical group de-rates. Stock's already given back 35% from the $637 high, still up 361% y/y. New PT is 51x CY27 P/E vs 67x before. Multiple compression, not thesis breakage.
Three things to watch into the Sept 4 print: backlog conversion against the supply overhang, 800G pluggable + DCOM/RLS-HyperRail ramp, and gross margin holding near 45% (last print 43%). Northland upgraded to Outperform on scale-optics TAM, Needham likes the Verizon dark fiber angle ($100M+ in FY27-28). AI infra names stay supply-constrained everywhere — Ciena's just not the only one.
Three cuts in one tape — Benchmark to $150 (Buy), Barclays to $96 (EW), Morgan Stanley to Underweight $80. That spread is the whole story: nobody knows what Baidu is worth when the legacy business is bleeding and the AI growth engine just hiccupped in front of everyone.
Q2 was the problem. Non-GAAP EPS $7.22 vs $9.84 expected, revenue $31.33B vs $31.95B. AI Cloud Infrastructure grew 50% YoY, but the SEQUENTIAL DECLINE is what broke the narrative — the market reads that as the AI story losing altitude. Ad revenue down EIGHT straight quarters, and management openly guided to more H2 weakness. Meanwhile capEx spiked and ERNIE investment is accelerating. Costs up, core down, AI lumpy. Horrid near-term mix, full stop.
The bull case isn't dead, just deferred. MS still pencils AI cloud infrastructure at 66% YoY growth, and the stock is down 34% in six months with RSI in oversold territory. That can produce a sharp technical bounce. But there's no catalyst in sight — ERNIE "back to top tier" is a 2027 story, and the market is paying for 2026 earnings today. Don't catch this knife; don't short it into oversold either. Let the rangebound chop sort itself out.
"The increased spending on foundational models, combined with a capital expenditure spike in the second quarter, raises concerns about near-term margin and earnings before incremental AI revenue materializes."
Exactly. Spend now, revenue later — and the market's patience for "later" is somewhere around zero at $91 a share.
BofA is holding Buy/$380 and the call is simple: Apple's next act is AI at the edge, and Ternus inherits a machine that's already printing value. Stock sits at $316.83 — up 41% on the year, still ~20% below BofA's number. The bull case isn't about the next iPhone. It's about the installed base monetizing through on-device intelligence.
The Cook math is staggering: roughly $4.5T of market cap created, ~$32M per hour, for nearly 15 years. Now a $4.62T company with 14% revenue growth and a 36x P/E. BofA's view is that the core doesn't change under Ternus — services, privacy, installed base growth stay the playbook. The upside comes from what sits on top: wearables (glasses, rings, AirPods with cameras), smart home, and personal assistants/robotics. All of it running on Apple silicon with local processing as the moat.
"By combining Apple silicon, on-device processing, privacy, operating systems, personal context, and a vast installed base, Ternus could cement Apple's device dominance in an AI-enabled era."
Bear case still exists — KeyBanc's stuck at Underweight/$250 on valuation, and the German ATT consent tweak ends a multi-year probe with a minor operational change. Nobody's selling AAPL over a consent prompt. The more interesting tell: Apple reportedly in talks with publishers for multiyear Siri content licensing, launch later this year. That's the AI narrative getting real, not theoretical. BofA's 20% upside case rests on Apple actually shipping the edge-AI story, not just talking about it.
TD Cowen lifts WDAY to $220 from $155 but stays at Hold — this is a "less bad, not more good" setup into the Aug 27 print. Checks are genuinely mixed: one global SI calls the quarter and outlook in-line, while a contracting partner flags real pricing pressure. That tension is the whole story.
"Higher discounting and elevated levels of contract complexity leading to longer close cycles, though resulting in longer contract terms."
So the trade-off is ugly on the surface — discounts compress near-term revenue, but longer terms smooth the cRPO path. TD Cowen sees modest upside to committed remaining performance obligations, little change to revenue. That's a margin story, not a growth story.
The real action is the Silver Lake overhang. The $198 tape is already pricing a take-out, not an earnings beat. Wells Fargo said it outright: PE deal discussions matter more than the quarter. Barclays at OW $200 and BMO at $182 are playing the margin/agentic AI angle; DB just downgraded to Hold at $220 after the run; UBS at Neutral $220 sees stable demand. Wide PT dispersion ($182-$220) tells you nobody has conviction on the operating metrics — everyone's waiting on the buyer.
Risk/reward is binary: either Silver Lake lands and the stock gaps to the mid-$200s, or it doesn't and you're long a decelerating software name with rising discounting at 6x forward sales. Hold is the honest rating here.
Rosenblatt starts AMZN Buy with a $335 PT. Not the most aggressive number on the Street, but the thesis matters more than the price target: AWS is the AI compounder and the market is still underpricing the exit rate.
AWS grew 37% in Q2 — fifth consecutive quarter of acceleration. Rosenblatt sees AWS EXITING 2026 AT 45% GROWTH vs. the Street's 38%. That's a 700bp gap at the most important inflection point in the story. Their math gets AWS to $335B+ by 2028.
The best part: AI spend isn't cannibalizing core cloud. Customers ramping AI are simultaneously growing base AWS usage — that's compounding, not substitution.
"We view Amazon as one of the most resilient leaders in our coverage, combining owned compute infrastructure with a physical fulfillment moat."
That's the whole bull case in one line — the only mega-cap with both the silicon/DC layer AND the last-mile logistics moat. 23 analysts just revised estimates upward, and the $18B Louisiana data center expansion shows capex is heading where the thesis says it should. Valuation is the only bogey — at $335 you're paying for that 45% exit rate to hold. Fair r/r into an AI buildout that's still early.
UBS pushed KEYS PT to $440 from $420 — and the real signal is buried in the estimate revisions: FY26/FY27/FY28 EPS goes to $11.47/$14.82/$17.10 from $10.11/$12.61/$14.50. That is a 13-18% jump across the model, not a target nudge. Buy stays.
Q3 came in clean: revenue $1.85B, +37% y/y and ~$110M above consensus; EPS $3.07 versus the $2.48 bogey. Orders cleared $2B for the second straight quarter, +56% y/y, and op margins improved ~400bps sequentially. Demand is broad-based AI. Management is actively fixing supply via long-term agreements and product redesigns — that's potential upside to Q4 and FY27, not just noise.
Q4 guide is the one thing PMs will parse: revenue $1.93-1.95B. That's above the $1.82B consensus, but below some whispers north of $2B. Fade the whisper. The guide still implies ~37% y/y growth and EPS of $3.34-3.40 against a $2.68 consensus. Not a demand problem; it's supply capacity catching up.
13 analysts revised EPS higher after the print — this is broad, not one loyalist. InvestingPro's fair value flags "overvalued," but static valuation models are late to rate-of-change stories like this. Watch the order trajectory, not the model.
Bull case: AI demand spans segments, supply fixes unlock revenue, and margin synergy from integration keeps printing. The new PT sits at 26x FY28 EPS — the market is paying up for visibility.
Bear case: a 56% order comp is a high hurdle. If orders flatten or AI capex pauses, the multiple compresses hard — and after a 109% y/y tape, KEYS is already pricing in a lot.
Neutral into the print, and the lone pre-earnings voice isn't fighting it. Guggenheim reiterates Neutral — not a bear call, more of a "the setup isn't easy" from a valuation and estimate perspective. GUGE SEES MINIMAL UPSIDE TO Q2/Q3 CONSENSUS ARR, and consensus already implies New ARR growth accelerating off increasingly difficult 2H comps. Stock at $201.63, down 10.6% on the week but still +72% YTD. The multiple is the issue: 34x EV/NTM recurring revenue. That's a full price for "in-line" risk.
The other brokers are louder and more constructive. Benchmark to $250, Cantor to $250, Wells Fargo to $230, Stifel Buy at $230 — the Stifel note flags resellers trending above plan. Guggenheim's own field checks back that up: all five partners met or beat quarterly expectations with healthy pipelines. So this isn't a demand problem. It's a "what's left for the stock" problem.
"CrowdStrike’s numerical setup does not appear easy, though it aligns with the company’s guidance."
Bull case stays intact on the AI angle — more AI means more malware proliferation, and CrowdStrike is the platform vendors pile into. Bear case is simpler: you're paying 34x recurring revenue for a setup where even the bulls are only modeling in-line guidance. CTO departure to launch an AI-focused security fund is a headline wrinkle, not a thesis changer. Earnings on the 26th. The field checks are encouraging; the chart and the multiple suggest expectations are already there.
Barclays initiates OW, $15 PT — ~56% above the $9.63 tape. Thesis: Bitdeer's AI/HPC infrastructure pivot is the right use of its ~3GW of powered land, and the Tydal colocation deal makes it concrete rather than slideware. 121MW signed at ~$160/kW/month, 3% annual escalators, 16 YEARS — $4.7B TOTAL CONTRACT VALUE. That's a real business, not a narrative.
Stock still sits ~40% BELOW LATE-JUNE. The decline comes down to two things: no revenue catalyst until early Q1 2027 and the $457M ATM equity raise in Q2 to fund capex. PMs hate paying today for 2027 earnings. Fair.
Q2 was noisy — EPS -$0.37 vs -$0.32 est, but revenue +47% YoY to $228.8M with AI Cloud growing. Benchmark holds $22 Buy; Needham trims $22→$20 on a weaker pipeline — just 47MW left at Tydal, Rockdale on hold, Clarington stays mining. That's the honest bear case: 121MW signed against ~1.9GW of stated capacity is a rounding error. The bull case is the land bank plus 16-year contracted cash flows, at a 2.5-beta price.
Barclays' PT says the post-deal selloff overshot. Needham's trim says don't chase. Both are right — until colo revenue actually prints in early Q1 2027, this trades on headlines and beta.
VERDICT: EVERYONE'S BEEN WAITING FOR THE REACCELERATION NARRATIVE TO GET ITS PULSE BACK — AND THE SETUP INTO NEXT WEEK'S PRINT IS THE CHEAPEST IT'S BEEN IN A WHILE. OPPENHEIMER REITERATES OUTPERFORM/$250, BUT THE REAL TELL IS THE cRPO GUIDE, NOT THE REVENUE NUMBER.
Stock at $206 (24.1x P/E, 0.6x PEG) — that's a value multiple on a growth re-rating story. The whole Street cluster is buys with PTs at $222-250 (Monness, JPMorgan, Stifel, Oppenheimer all in the same camp). Not a lot of dispersion. That's fine when the bar is this low.
Oppenheimer's core thesis: partner checks show strong net new Agentforce and Data Cloud ARR, and 77.64% gross margins give CRM room to fund the AI push without wrecking the P&L. Outsize capital returns keep the value crowd anchored while the growth story catches up.
The line to watch on the call: IF MGMT GUIDES 13-14% Q3 cRPO GROWTH, THAT CONFIRMS THE REACCELERATION THESIS FOR 2H FY27 — and sets up multiple expansion into Dreamforce next month. That's the investor day catalyst nobody's pricing in yet.
"If management guides 13% to 14% for third-quarter current remaining performance obligation growth, it will support the revenue reacceleration narrative expected in the second half of fiscal 2027."
Not a barn-burner take, but the r/r is skewed positive. Low bar + AI narrative + buybacks + event-driven catalyst = the kind of asymmetric setup PMs should have on the radar into the print.
Barclays upgrades CCI to Overweight while CUTTING the PT to $84 from $92 — that's a conviction call, not a catch-up. Lynch says the Street is wrong on satellite risk: Starlink's terrestrial ambitions are a net positive for towers, not an existential threat. That's the bull case in one line.
"Starlink's desire for a terrestrial network is a net positive for towers."
The bear narrative collapses into one bogey: the services gross margin guidance cut for 2026. Barclays says that's noise — services and leasing have historically had weak correlation. They model core leasing ACCELERATING to 2.0% in 2027 from 1.7% in 2026, driving a 4.2% organic contribution to site rental billings next year. At 15.0x 2027 AFFO (6.7% AFFO yield) with a well-covered 5.7% dividend, the return profile beats most of the REIT universe.
Q2 was a beat (EPS $0.69 vs $0.39 est, revenue $1.01B vs $995.97M). But post-print PT cuts kept rolling — Bernstein to $96, Truist to $87, Citizens to $120. The Street is split: Barclays sees a simpler, de-levered CCI (net debt down to 6.3x after the fiber sale) re-rating from here; the skeptics still think the tower leasing recovery is too slow. The upgrade with a lower PT tells you the multiple, not the model, is where the upside lives.
Barclays finally gets it. Upgraded AMT to Overweight from Equalweight, PT to $198 (from $195). The call is growth, not multiple — and the numbers back it up. Barclays models 2.6% domestic colo/amendment growth in 2026, vs 1.7% for Crown Castle and 2.0% for SBA. The spread holds in 2027 too: 2.3% vs 2.0%.
The whole bull case hinges on AT&T. They've only upgraded 50% of 4G sites to 5G — trailing both T-Mobile and Verizon — and they've rolled off the master lease with AMT. That's a massive catch-up cycle sitting in AMT's lap.
"The pace of AT&T's deployment in 2027 represents the biggest swing factor in American Tower's domestic revenue next year."
International adds a second gear: 2.7% growth in 2026, accelerating to 5.0% in 2027, on Europe/Africa exposure and CPI escalators. CoreSite is the third: Barclays models 15% YoY revenue growth in 2027 on pricing power, new capacity, and incremental sales off the existing power base. The Q2 print (adj EPS $1.86, rev $2.75B) beat and the company raised FY guidance a second time — so the momentum is real.
Street's moving in: Bernstein/SocGen at $214, Scotiabank at $220, BMO at $195. The low PEG (0.15 at 23.9x P/E) tells you the market still isn't paying for the acceleration. We'd watch the Texas data center audit — small overhang, but CoreSite has exposure.
TD Cowen's bump to $48 from $45 isn't the story. The read on headcount is. They're looking at aggressive hiring as a pipeline tell, not a margin drag — a contrarian take on opex that most PMs should care about heading into the Sept 3 print.
STOCK'S ALREADY +46% OVER SIX MONTHS. So this is momentum, not value. TD Cowen expects a beat + FY27 guide raise. Company's TTM average beat is ~5% — familiar bogey, bar set where Samsara usually clears it.
Bull case: new product adoption showing up in the quarter, platform ROI stories landing, and the physical AI angle (attacking labor budgets, not just fleet tracking) opening a bigger TAM. Management added bodies aggressively — that's the tell that pipeline visibility is real, not hope.
"The company's aggressive headcount additions point to strength in pipeline opportunities."
Bear case rests on Piper Sandler's downgrade to Neutral ($40 PT) on app download declines. That's consumer-grade telemetry for an enterprise IoT platform with a 3p sales motion. Not sure we can read much into that one. Wolfe (Outperform), Guggenheim (Buy, $45), KeyBanc (Overweight) all sit on the other side post Q1 and the Beyond conference.
Net: the smart money's circling a beat-and-raise. Physical AI gives this a FY27 story, not just a Q2 print. Momentum's extended, but the setup into Sept 3 is constructive. Longs need that guide raise to do the heavy lifting — TD Cowen clearly thinks it's coming.
Verdict: Bearish sentiment + low bar = beat-and-raise setup into the Aug 26 print. Benchmark reiterates Buy/$570 while the stock sits at $403 (down 34% on the year, $77B cap). That's a lot of pain already in the tape, and the analyst floor sees above-consensus results and stronger 2H FY26 guidance.
"Investor sentiment appears relatively bearish, setting a low bar for results."
Core thesis: EDA/sim/IP markets accelerated YoY, RPO/backlog inflecting — Benchmark models ~$12B RPO exiting FY26, a QoQ improvement. Gross margin at 82.6% provides the downside cushion. Nine analysts revised FY estimates up recently. The set-up is genuinely good.
The bulls are getting louder. Piper Sandler upgraded to Overweight/$550 on Intel recovery (SNPS's largest customer, finally showing life). KeyBanc stays at $600. S&P moved the outlook to positive after SNPS repaid $4.3B of the $14.3B Ansys debt — helped by Nvidia's $2B investment. The bear case? The chip manufacturing software exit (Samsung, SK Hynix affected) is a messy transition, and nobody gets FY27 guidance until the Sept 30 analyst day — the first full one in 2.5 years. The Aug 26 call is the near-term bogey; the analyst day is the multiple-expansion event. If they show the AI design narrative working there, $570 starts to look conservative.
Rosenblatt keeps Buy, PT to $15 from $13. Not the headline — the estimate move is. New 2027 adj EBITDA est $381M vs $317M prior, and CONSENSUS WAS $244M PRE-PRINT. That's a 56% gap. PT sits at 22x (down from 23x) on a much bigger base — multiple compresses because the number explodes.
Thesis: the pattern day trader rule removal is structural, not a volume sugar rush. Webull sees mix shift to higher-yield margin trades and materially smaller trade sizes — traders no longer constraining themselves means more frequent activity and better revenue capture per unit of flow. Q2 confirms it: revenue $198.8M (+51% YoY), record adj op profit $62.6M (+169% YoY). LTM growth 45% with 76% gross margins — this is a quality fintech model, not a commoditized brokerage.
"The PDT rule change will drive accelerating revenue growth beyond just trading volume."
Rosenblatt flags Q3 2026 trends already encouraging, new product revenue disclosure, and international opportunities sitting entirely outside estimates. That's the free option. Risk: 22x is a full multiple for a trading-dependent name if volumes normalize — but the sequential capture-rate improvement argues the model just got better structurally, not bigger cyclically. The $0.05 EPS print tells you less than the operating leverage story does.
Piper Sandler takes the baton at $131 (from $126, Overweight) — the high end of the street with RBC at $120 — and the whole call rests on one thing: the bookings inflection is finally converting to a revenue story.
Q4 BOOKINGS +93.1% YoY. Total backlog $1.945B (+38.4%). NTM backlog $996M. That's ~92% visibility into FY27 revenue guidance of "approaching $1.1B." You don't need to squint hard to see why Piper is leaning in.
The sour note: EPS $0.37 missed by a penny (revenue $289.8M beat, +6.1% organic). Margins are the problem — specifically, the last of the low-margin legacy backlog burning off. Piper frames it as temporary, concentrated in Q1/H1 FY27 and fading through the year, with Mercury hitting the low end of its EBITDA target range by FY27 year-end.
The kicker is munitions program awards converting over the next few quarters. If those long-term contracts hit, this goes from a visibility story to an acceleration story. Stock's already up 63.5% in a year — so the easy money is made. But with 92% visibility and the margin drag set to decay, the r/r still tilts bullish from here.
Needham pushes PT to $21 from $18, keeps Buy. Stock sits at $18.31 — RIGHT AT the 52-week high ($18.70) after a 62% six-month rip. Momentum is real and the street keeps chasing it.
The thesis is clean: AI-native solutions are scaling, pulling incremental marketing and R&D budgets from FMCG customers. Needham thinks the market still underappreciates this — and the setup is a re-rating as that AI revenue compounds.
"The firm believes the market underappreciates NIQ’s AI-native capabilities and expects a favorable re-rating as revenue from these solutions grows."
Q2 backed it up. Adjusted EPS $0.27 vs $0.21 consensus, revenue $1.12B, and a raise to FY guidance — FIFTH straight quarter of beating across key metrics. Seven analysts revised numbers up since the print.
Buyback optionality is the sleeper. $506M levered FCF, cleaner balance sheet — Needham sees authorization within 12 months. That's a nice torque at these levels.
The sell-side cluster is consolidating around one story: Needham's gone $12→$18→$21 since the quarter, BMO sits at $20, Stifel at $17. Everyone's on the same page — AI is the engine, and at $18.31 there's still room to the highest targets.
Tigress raises to $130 (Buy maintained) on AI-infrastructure demand, record backlog, accelerating earnings growth. BTIG cuts from $120 to $100 on margins. Stock sits at $64.91. That's the whole debate in one line: revenue machine vs. earnings mess.
Q2 was the full contradiction. REVENUE JUMPED 111% TO $1.262B (organic plus the Bowers acquisition). ADJUSTED EBITDA ~$155M — crushed the $123M consensus AND the company's guidance. Then the air came out: ADJUSTED EPS -$0.37 vs +$0.21 expected. That's a $0.58 miss. Not a rounding error.
"Legence's AI infrastructure and energy efficiency creating a long-term growth platform." — Tigress
Bull case: AI-infra scale, operating leverage, capital efficiency → rising ROIC and economic profit. Capital allocation targets rapid deleveraging, strategic M&A, fabrication expansion. Six analysts bumped forward estimates. If the AI supercycle is real, LGN is a leveraged way to play it.
Bear case: the EPS miss says margins aren't there yet, and BTIG cutting the PT despite an EBITDA beat tells you analysts are watching the same thing. The raised full-year outlook got buried under the negative print. The tape says "show me."
Tigress' $130 vs. the $64.91 tape is a massive spread. Either the bull case is way early, or the market is pricing execution risk bulls ignore. For PMs: this is a high-beta, high-conviction AI-infra proxy with a messy P&L — you're trading the trajectory, not the multiple. Size it like a special sit.
AAOI — US-made optical modules carry a real premium now. Hyperscalers are paying for supply-chain assurance, not just specs. That's a geo-arbitrage tailwind for AAOI's US footprint. The recent optics pullback reads as a re-entry, not a thesis break.
AMD — Net mixed. Bear: Intel can hold or increase its TSMC allocation to cap AMD's server CPU ramp — market looks over-optimistic on foundry share. Bull: Helios roadmap is credible — 4x AI energy efficiency already delivered vs a 3x target, MI455X, UALink, ROCm. NVIDIA accepting AMD PRs into NIXL is small but real — software moat erosion. Custom-silicon chatter (OpenAI/Meta getting ~10% "almost-free" shares) supports pipeline but adds dilution.
AMAT — Non-litho etch/dep lead times out to 24 months. WFE forecasts keep expanding. AMAT is a prime advanced-node + HBM capacity beneficiary. Supply constraints are protecting pricing.
ANET — Keysight confirms 1.6T optics ramping now, demand FAR exceeding supply. ANET has live 1.6T exposure — estimate risk skews positive. But supply stays the governor for several quarters. Demand-positive, supply-negative setup.
APH — AI premiumizes high-speed interconnect, advanced PCBs, passives. Higher frequencies and tighter tolerances raise qualification barriers — customers consolidate with proven suppliers. APH wins on both content growth and vendor consolidation.
ASML — Advanced litho lead times at two years. ASML expected to print ~EUR 21B FCF in 2028. Multi-year, supply-constrained tailwind.
ATEYY — Test intensity rises faster than wafer volume. HBM, chiplets, silicon photonics all need longer test times, more insertions, system-level test. Advantest monetizes the buildout regardless of which memory producer wins.
AVGO — Google's Marvell warrant deal breaks the TPU single-vendor assumption — AVGO broke ~5% when MRVL jumped 9%. Google is multi-sourcing across AVGO, MediaTek, Marvell. But Tomahawk 6 CPO is commercial, and AVGO is the critical packaging/photonics enabler. Negative on custom-silicon exclusivity, positive on CPO.
BABA — Profit -75% is a deliberate sacrifice for AI scale, layered on weak China consumer; Ant flat with AI/LLM spend up. Market split: strategic investment vs value trap. China model pricing is the structural wildcard — Moonshot/ZAI at ~1/4 of Claude cost pull BABA into the price war. Watch Cloud revenue growth and AI-margin trajectory.
BE — The speed-to-power window is real: within a 100GW/five-year DC demand frame, near-term grid share compresses to 50-60GW — everything Bloom can ship gets deployed. PA HB1834 lists fuel cells as clean firm energy without fuel restriction; Nebius Vineland's switch from gas gensets to solid oxide validates the substitution trade. NIMBY risk now quantifiable — 71% of Americans oppose local DCs. Bear case: ~10%/yr cost-down may not be enough if grid power gets abundant.
BHP — Copper is the AI-electrification long. BHP offers copper optionality inside a diversified major — iron ore and China are the drag. Debate is whether copper growth offsets that.
CLS — Keysight's SE Asia business more than doubled — validates the Thailand/Malaysia/Vietnam buildout. CLS has live 1.6T optics exposure. Supply constraints are the near-term risk, not demand.
COHR — 1.6T ramping, demand far exceeding supply. COHR has direct exposure — estimate risk skews positive. Key debate: can they add capacity fast enough for the 1.6T supercycle.
DXYZ — Anthropic net ARR adds re-accelerated to ~$16B/month, implying a >$200B run-rate by year-end. The $65B print likely reflects a shift from gross to net reporting — more conservative, not weaker. DXYZ is the public proxy for private AI/space. Any ARR slowdown is catastrophic for the IPO-skeptic framing.
FCX — Copper remains the AI-electrification long, strong YTD. No FCX-specific data in the source. The debate is supply vs AI/grid demand.
FLEX — Cerebras added manufacturing lines at Flex and expects capacity up more than 10x in 2026. Plus SE Asia EMS diversification. That's wafer-scale AI inference exposure.
FN — Keysight SE Asia more than doubled; Fabrinet is the most direct exposure to that geography. 1.6T ramping now. Positive estimate signal. Supply still the governor.
FORM — Probe card intensity rises with HBM and chiplet complexity. More insertions, more system-level test. FORM monetizes AI test complexity regardless of memory winner.
INTC — Razor Lake rumored on TSMC N2X, Nova Lake on N2P — a big 2nm outsourcing expansion, but cuts single-process risk. Underappreciated lever: Intel can hold/increase its TSMC allocation to block AMD's server CPU capacity. Contrarian 18A bull case persists — management hints at massive customer demand, Ohio moving fast, market late to Intel inflection points.
INTU — AI-native SaaS is stealing share. Rillet with fewer than 50 sellers is near 100% quota, sourcing 11% of wins from Sage Intacct and 8% from NetSuite — Intuit also a target. Early, but the rate of change is fast. If AI-native horizontal software scales, Intuit's SMB distribution moat erodes.
JBL — SE Asia buildout validation. Jabil benefits from customer diversification efforts across EMS. AI infrastructure growth, supply-constrained.
KLAC — Design rules, complex packaging, silicon photonics, HBM stacking — all require more inspection and process control. Intensity rises faster than wafer volume. WFE forecasts expanding. KLAC monetizes AI complexity.
LITE — 1.6T ramp confirmed. Direct exposure, estimate risk positive. Supply governor for several quarters. Can they ramp capacity fast enough?
LRCX — Etch/dep lead times at 24 months, WFE expanding. Advanced-node, HBM, 3D stacking all land here. Supply constraints protect pricing.
META — Custom-silicon chatter: OpenAI and Meta each get roughly 10% "almost-free" AMD shares. If true, AMD pays with equity for customer commitments and Meta broadens beyond NVIDIA. Low confidence — no financials in source.
MRNA — MRNA +177% on Phase 3 individualized mRNA cancer therapy: 49% lower recurrence/death risk in 1,137 melanoma patients. Every dose custom — tumor sequenced, HLA typed, AI predicts ~34 neoantigens. Model frozen pre-Phase 2 from 2021-era AI. Caveat: melanoma may be a special case; mRNA doesn't solve TME in other cancers.
MSFT — ~70% of Microsoft's AI revenue comes from OpenAI. That's concentration risk for the Azure AI narrative — open-weight model headwinds and pre-IPO accounting noise at OpenAI compound it. If OpenAI ARR reacceleration is real, MSFT is fine. If OpenAI loses share, MSFT is exposed. Needs non-OpenAI AI revenue to scale.
MU — Memory is a one-way trade up. Morgan Stanley sees DDR4 +50% Q/Q in Q3, more upside in Q4; global server 2Q missed on SUPPLY, not demand. HBM4 spec: >2.8 TB/s per stack, 8 stacks >22.4 TB/s. US-made DRAM commands a premium (Koreans responding with US fabs), and a major player is building a custom HBM/3D IC team — the next frontier is co-design, not just capacity.
NBIS — $5B converts add growth capital. PA removing DCs from Fast Track permitting raises the scarcity value of projects already advanced — Nebius Vineland. The gas genset → solid oxide shift confirms power constraints on the flagship build.
NOK — JPM sees ~100% upside. AI/cloud orders accumulated EUR 38B over two quarters, EUR 24B shippable within 12 months. 2Q26 IP Networks orders >40% of total, >EUR 1.1B. If pace holds, FY26 sustained orders could hit EUR 52B and FY27 sales land >100% above consensus.
NVDA — Mixed tape. BEAR: high-end CCL is the next bottleneck — 4-6 month lead times, Doosan sole supplier for Vera Rubin compute boards initially; EMC/Doosan equipment installs slipped ~6 months, TUC Thailand phase 2 delayed ~2 months. BULL: Spectrum-X Ethernet Photonics in production; MLPerf shows GB300 at 9,360 tokens/s/GPU vs H200's 1,113 — ~5x tokens at 2x watts. China H200 dribble (~10K chips, mostly stuck in Hong Kong) is modest. NIXL accepting AMD PRs is a small crack in the software moat.
ONTO — Advanced packaging, silicon photonics, HBM stacking drive process control demand. ONTO is a direct beneficiary. Test intensity rises with AI complexity.
ORCL — Project Jupiter hitting siting friction: New Mexico State Land Office rejected the pipeline ROW twice. Adds permitting lead-time risk to a flagship AI build. NetSuite also gets targeted by AI-native SaaS — Rillet pulling 8% of wins from it. Infrastructure friction delays backlog or just raises costs.
PHM — NVDA/SPAN/PulteGroup piloting ~100 new homes with a liquid-cooled 16-GPU Blackwell box, $0 upfront for homeowner. Immaterial near-term revenue. But a differentiation and marketing angle in select new communities.
PLTR — Karp's argument: proprietary data plus ontology beats frontier labs — the deployment layer is the moat, not the model. That's the bull case in an agentic world. Debate: does ontology lock-in survive open-weight commoditization?
RDDT — Reddit's negotiating position has deteriorated. Google's grounding gap closed, YouTube citation share overtook Reddit, and Google News/Merchant Center/Places cover most of what Reddit was a shortcut to. Deal likely but on reset terms, not the old $60M licensing scale. Key variable: does prompt-level data keep flowing to Google?
RVI — Anthropic net ARR adds re-accelerated to ~$16B/month — >$200B run-rate by year-end. $65B print likely the gross→net reporting shift. RVI is the public Anthropic proxy. IPO-skeptic framing requires massive monthly adds to sustain the capex trade.
SANM — Cerebras added lines at Sanmina — capacity up more than 10x in 2026. CS-4 modularity cuts install time from days to hours. New AI systems manufacturing opportunity.
SCCO — Cautious posture. Demanding valuation, $20.5B Tia Maria capital program needs flawless execution. Production grade decline, leadership transition, governance concentration flagged. Prefer simpler copper beta like TECK.
SMIC — Q2 revenue just over $3B, +19.7% QoQ, +36% YoY — beat the midpoint handily. Management says AI companion chip demand is enormous, pulled forward shipments ~40%. Self-sufficiency plus AI demand accelerating. Sustainability of the pull-in is the open question.
SNDK — NAND tightness plus AI infra demand. No company-specific data point in source. YMTC surpassing Kioxia in NAND bit shipments is a China share-gain watch item.
SSNLF — Foundry prices moving UP 10-15% — confidence in AI accelerator demand. Taylor Fab 2 accelerating, SEMI certs pulled early, 2nm mass production targeted 2030, Tesla widely cited as anchor. Shareholder return plan north of KRW 100T floors the stock but tensions with capex. Process gap to TSMC remains: SF2P/SF2P+ next year, but 1.4nm slips to 2029 vs A14 in 2028.
TECK — Simpler copper beta than SCCO, no governance overhang. Strong YTD. Debate is copper supply vs AI/grid demand.
TER — Test intensity rises with HBM, chiplets, silicon photonics. TER monetizes the buildout regardless of memory winner.
TSLA — Samsung Taylor Fab 2 targeting 2nm, Tesla widely cited as key customer — KRW 22.76T contract manufacturing agreement in 3Q25. Construction momentum accelerating. Gives Tesla a US leading-edge foundry partner outside TSMC. Debate: is Tesla's custom silicon volume big enough to fill the fab?
TSM — Critical packaging and silicon-photonics enabler for switch-level CPO — NVDA Spectrum-X and AVGO Tomahawk 6 both depend on TSMC packaging. A14 mass production in 2028 while Samsung 1.4nm slips to 2029. Global Unichip raising $2.04B to front-run packaging capacity for ASIC clients — custom silicon volume flows straight to TSM.
TTMI — AI premiumizing high-performance multilayer PCBs and materials. Qualification barriers rise. Vendor consolidation benefits TTMI.
VIAV — Negative near-term supply constraint risk. 1.6T in volume production but supply governs. Demand strong, revenue conversion capped by upstream components.
WMT — Print tonight — key US consumer read-through. No numbers yet. Confirmation of softness feeds the AI capex financing cost and discretionary pressure argument; a surprise relieves risk appetite.
DELL — No fresh signals in the feed. AI server demand strong but supply-constrained — same memory/CPU/substrate bottleneck. The 1.6T and CCL pinches feed into server cost pressure.
KLAR — No fresh signals. Consumer credit/BNPL sits directly in tonight's WMT read-through path. If Walmart guides soft, charge-offs become the tell.
KEEL — Nothing on the tape. No signal, no position — this stays on the bench until we get a real data point.
CEG — No company-specific data, but the power scarcity theme is the trade. Denmark made DCs the lowest grid priority; behind-the-meter systems are showing turbine failures and margin risk. That's a relative positive for regulated and nuclear generation.
SPSC — No fresh signals. Pop Mart's warning is a China consumer yellow flag; tonight's WMT print sets the US tone. Retail SaaS follows both.
ZS — No fresh signals. AI agents expand the attack surface and identity sprawl — a secular tailwind for zero trust. But no estimate catalyst in the feed.
JKHY — No fresh signals. Bank IT budgets are a lagging indicator; consumer health via WMT is the near-term driver.
GPN — No fresh signals. Payments is the cleanest consumer-spending tape after WMT. Guide soft, estimate risk skews down.
RBRK — No fresh signals. Cyber stays a budget-safe line item; AI agent traffic is the new vector. No catalyst here.
MDB — No fresh signals. AI apps are the demand driver, but open-weight models compress the value layer. Watch the AI workload mix.